Only run asserts in debug mode

[LilithHafner]

This is motivated by the ease of performing the optimization, not by profile guided optimization which is a huge red flag.

[github-actions]

Benchmark Results

	e04e15c...	4756eb9...	e04e15c.../4756eb9021ba5b...
constructor n=100 σ=0.1	9.49 ± 3.2 μs	10.9 ± 1.3 μs	0.87
constructor n=100 σ=1.0	9.43 ± 0.89 μs	12.9 ± 2.9 μs	0.729
constructor n=100 σ=10.0	9.47 ± 0.46 μs	11.3 ± 1 μs	0.838
constructor n=100 σ=100.0	10.1 ± 3.4 μs	10.1 ± 2.9 μs	1.01
constructor n=1000 σ=0.1	0.0569 ± 0.0026 ms	0.0579 ± 0.0038 ms	0.983
constructor n=1000 σ=1.0	0.0591 ± 0.0028 ms	0.0593 ± 0.004 ms	0.996
constructor n=1000 σ=10.0	0.0647 ± 0.0018 ms	0.075 ± 0.0019 ms	0.863
constructor n=1000 σ=100.0	0.0727 ± 0.0019 ms	0.0792 ± 0.0026 ms	0.917
constructor n=10000 σ=0.1	0.58 ± 0.11 ms	0.61 ± 0.05 ms	0.951
constructor n=10000 σ=1.0	0.919 ± 0.41 ms	0.626 ± 0.029 ms	1.47
constructor n=10000 σ=10.0	1.02 ± 0.53 ms	0.648 ± 0.28 ms	1.57
constructor n=10000 σ=100.0	0.74 ± 0.1 ms	0.813 ± 0.057 ms	0.911
delete ∘ rand n=100 σ=0.1	5.59 ± 0.21 μs	5.79 ± 0.23 μs	0.965
delete ∘ rand n=100 σ=1.0	5.9 ± 0.21 μs	6.01 ± 0.23 μs	0.982
delete ∘ rand n=100 σ=10.0	7.18 ± 0.43 μs	7.19 ± 0.43 μs	0.999
delete ∘ rand n=100 σ=100.0	16.4 ± 0.86 μs	15.8 ± 0.86 μs	1.04
delete ∘ rand n=1000 σ=0.1	0.0557 ± 0.0012 ms	0.0575 ± 0.0012 ms	0.969
delete ∘ rand n=1000 σ=1.0	0.0604 ± 0.005 ms	0.0614 ± 0.0012 ms	0.984
delete ∘ rand n=1000 σ=10.0	0.0626 ± 0.001 ms	0.0635 ± 0.001 ms	0.986
delete ∘ rand n=1000 σ=100.0	0.0866 ± 0.0018 ms	0.0845 ± 0.0014 ms	1.02
delete ∘ rand n=10000 σ=0.1	0.62 ± 0.021 ms	0.611 ± 0.015 ms	1.01
delete ∘ rand n=10000 σ=1.0	0.646 ± 0.014 ms	0.654 ± 0.013 ms	0.986
delete ∘ rand n=10000 σ=10.0	0.651 ± 0.013 ms	0.66 ± 0.018 ms	0.987
delete ∘ rand n=10000 σ=100.0	0.666 ± 0.012 ms	0.671 ± 0.014 ms	0.993
empty constructor	3.61 ± 1.3 μs	4.25 ± 1.1 μs	0.849
intermixed_h n=100 σ=0.1	14.2 ± 1.7 μs	16.1 ± 2.5 μs	0.88
intermixed_h n=100 σ=1.0	14.6 ± 1.8 μs	15.4 ± 1.9 μs	0.952
intermixed_h n=100 σ=10.0	18 ± 3.8 μs	16.9 ± 3.7 μs	1.07
intermixed_h n=100 σ=100.0	17.1 ± 2.3 μs	18.6 ± 2.8 μs	0.919
intermixed_h n=1000 σ=0.1	0.126 ± 0.01 ms	0.126 ± 0.0096 ms	0.993
intermixed_h n=1000 σ=1.0	0.127 ± 0.013 ms	0.133 ± 0.0095 ms	0.952
intermixed_h n=1000 σ=10.0	0.132 ± 0.013 ms	0.134 ± 0.013 ms	0.986
intermixed_h n=1000 σ=100.0	0.143 ± 0.018 ms	0.151 ± 0.019 ms	0.951
intermixed_h n=10000 σ=0.1	1.44 ± 0.28 ms	1.37 ± 0.18 ms	1.05
intermixed_h n=10000 σ=1.0	1.38 ± 0.33 ms	1.41 ± 0.2 ms	0.982
intermixed_h n=10000 σ=10.0	1.53 ± 0.45 ms	1.52 ± 0.48 ms	1.01
intermixed_h n=10000 σ=100.0	1.61 ± 0.43 ms	1.41 ± 0.21 ms	1.14
pathological 1	0.363 ± 0.0029 μs	0.37 ± 0.0024 μs	0.981
pathological 2	0.952 ± 0.0045 μs	0.949 ± 0.0049 μs	1
pathological 3	20.8 ± 1 ns	20 ± 0.71 ns	1.04
pathological 4	1.03 ± 0.0054 μs	1.04 ± 0.0039 μs	0.986
sample n=100 σ=0.1	30.6 ± 0.81 ns	29.7 ± 1 ns	1.03
sample n=100 σ=1.0	0.034 ± 0.0022 μs	0.0339 ± 0.0022 μs	1
sample n=100 σ=10.0	23.6 ± 5.6 ns	23 ± 5.5 ns	1.02
sample n=100 σ=100.0	20.7 ± 5 ns	20 ± 4.2 ns	1.04
sample n=1000 σ=0.1	28.6 ± 5.1 ns	27.8 ± 5.7 ns	1.03
sample n=1000 σ=1.0	0.0368 ± 0.0021 μs	0.0371 ± 0.0021 μs	0.993
sample n=1000 σ=10.0	25.1 ± 5.6 ns	24.2 ± 6.3 ns	1.04
sample n=1000 σ=100.0	21 ± 4.9 ns	20 ± 4.2 ns	1.05
sample n=10000 σ=0.1	0.0373 ± 0.0011 μs	0.0369 ± 0.0012 μs	1.01
sample n=10000 σ=1.0	0.04 ± 0.0017 μs	0.0403 ± 0.0014 μs	0.993
sample n=10000 σ=10.0	25.7 ± 5.9 ns	24.2 ± 6.8 ns	1.06
sample n=10000 σ=100.0	20.8 ± 4.4 ns	20.5 ± 4 ns	1.01
update ∘ rand n=100 σ=0.1	0.103 ± 0.0023 μs	0.106 ± 0.0025 μs	0.971
update ∘ rand n=100 σ=1.0	0.11 ± 0.0029 μs	0.112 ± 0.0028 μs	0.974
update ∘ rand n=100 σ=10.0	0.137 ± 0.02 μs	0.141 ± 0.018 μs	0.969
update ∘ rand n=100 σ=100.0	0.346 ± 0.037 μs	0.324 ± 0.033 μs	1.07
update ∘ rand n=1000 σ=0.1	0.104 ± 0.0029 μs	0.106 ± 0.0028 μs	0.977
update ∘ rand n=1000 σ=1.0	0.11 ± 0.0031 μs	0.113 ± 0.0031 μs	0.978
update ∘ rand n=1000 σ=10.0	0.13 ± 0.0063 μs	0.134 ± 0.0064 μs	0.97
update ∘ rand n=1000 σ=100.0	0.343 ± 0.016 μs	0.349 ± 0.017 μs	0.982
update ∘ rand n=10000 σ=0.1	0.112 ± 0.002 μs	0.113 ± 0.0021 μs	0.983
update ∘ rand n=10000 σ=1.0	0.118 ± 0.0039 μs	0.118 ± 0.0028 μs	0.999
update ∘ rand n=10000 σ=10.0	0.127 ± 0.0024 μs	0.128 ± 0.0023 μs	0.993
update ∘ rand n=10000 σ=100.0	0.309 ± 0.0036 μs	0.316 ± 0.0048 μs	0.978
time_to_load	0.725 ± 0.019 s	0.728 ± 0.027 s	0.995

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

[LilithHafner]

I can't reproduce the 1.57x "constructor n=10000 σ=10.0" improvement. Locally there is no performance improvement at all.